A Probabilistic Method for Software Product Measurement

In controlling the software development process there is a need to discover how a software product will evolve and to monitor this evolution. Measures of internal product attributes can provide the necessary objective data from which the software developer can gain this important engineering information. Outlier analysis has been proposed as a technique to interpret such measurement data. However it is handicapped through a lack of a theoretical understanding of how measures interact. In this dissertation a probabilistic method is proposed which helps formulate the relationship between internal software product measures. The method provides a way for sampling the distribution of measurement values for particular models of software products. The reward of such an approach is that a greater understanding of the outlier analysis technique is achieved. The question of what constitutes an outlier is formally analysed. The method is applied to the measurement of predicate expressions in Z specifications and involves the design of some new algorithms.